Pearson Correlation in SPSS: Procedure, Output, and Reporting

SPSS runs a Pearson correlation through the Bivariate Correlations procedure, which produces a compact but easy-to-misread output table: three rows per variable pair, a symmetric matrix, and an N that can quietly change from one cell to the next. This guide covers the menu path, the assumption checks to run before you trust the number, how to read every row of the output table, and the exact line to write up in APA format. For the underlying statistic itself — what it measures, and when to prefer Spearman’s rho or Kendall’s tau instead — see the correlation coefficient guide; this page covers the SPSS procedure specifically.

Running the Bivariate Correlations procedure

Go to Analyze > Correlate > Bivariate. Move the two (or more) continuous variables you want correlated into the Variables box — SPSS will correlate every possible pair if you enter more than two, producing a full matrix rather than a single value. (New to the software generally, or deciding whether it’s the right tool for your project? See what SPSS is and who uses it and SPSS vs. R first.)

• Correlation Coefficients: check Pearson. The dialog also offers Kendall’s tau-b and Spearman in the same box — leave those unchecked unless you deliberately want a second, non-parametric coefficient run alongside it for comparison.
• Test of Significance: Two-tailed is the default and the correct choice unless your hypothesis specifies the direction of the relationship in advance; a one-tailed test roughly halves the reported p-value for the same data, so switching it after seeing the result is a form of p-hacking, not a legitimate analytic choice.
• Flag significant correlations: leave this checked. It adds an asterisk convention to the output matrix (see the output-table section below) that makes a large matrix scannable at a glance.
• Options: this is where Missing Values is set to Exclude cases pairwise or Exclude cases listwise — covered on its own below, because it changes what N means in your output. The Options dialog also lets you add means, standard deviations, and cross-product deviations/covariances to the output if you want them, though most write-ups only need the correlation matrix itself.

Check linearity and outliers before you trust the number

Pearson’s r only measures the strength of a linear relationship, and a single extreme point can pull it substantially in either direction — the same assumption that interval/ratio-level data makes possible in the first place. Run this check before the Bivariate Correlations procedure, not after, so a bad r doesn’t get explained away instead of diagnosed:

• Produce a scatterplot via Graphs > Chart Builder (choose Scatter/Dot, drag the two variables onto the axes) or the legacy path Graphs > Legacy Dialogs > Scatter/Dot > Simple Scatter.
• Linearity: the points should trace a roughly straight-line pattern. A clear curve (a relationship that rises then falls, or accelerates) means Pearson’s r will understate the true strength of association, sometimes badly — a curved relationship can produce an r near zero even when the two variables are strongly, systematically related. Consider a nonlinear model or a monotonic-only measure like Spearman’s rho if the curve is not correctable by a transformation.
• Outliers: look for one or a few points sitting well away from the main cloud. Because r is computed from products of deviations from the mean, a single outlier — especially one at the extreme of either variable’s range — can inflate or deflate the coefficient disproportionately to its share of the sample. Add case labels in the Chart Builder (or double-click the chart in the output viewer) to identify which case it is before deciding whether to investigate, transform, or exclude it; exclusion needs a documented, principled reason, not just “it made the correlation stronger.”

Both checks are about whether Pearson’s r is the right summary at all, not just whether the number that comes out is trustworthy — a nonlinear or outlier-driven pattern can produce a reportable-looking r that misdescribes the actual relationship.

Reading the Correlations output table

SPSS prints a matrix with one row-block per variable and one column per variable; each cell where a row and column meet contains three stacked values:

• Pearson Correlation: the coefficient itself, from −1 to +1. The diagonal (a variable correlated with itself) is always exactly 1.000, with no significance value beneath it — that’s expected, not an error.
• Sig. (2-tailed): the p-value for the test of whether the true correlation differs from zero. This is a test of the coefficient’s distinguishability from zero, not a test of how large or meaningful it is — a very small, practically trivial r can still show a tiny p-value in a large sample, and a moderate r can fail to reach significance in a small one.
• N: the number of cases actually used to compute that specific cell. In a matrix with more than two variables and any missing data, this number can differ from cell to cell — see pairwise vs. listwise deletion below for why.

With Flag significant correlations checked, SPSS marks the Pearson Correlation value with a single asterisk (*) for p < .05 and a double asterisk (**) for p < .01, and prints a footnote explaining the convention beneath the table — useful for scanning a large matrix, but always report the exact p-value in your write-up rather than just the asterisk.

Listwise vs. pairwise exclusion, and why it changes N

Set under Options > Missing Values:

• Exclude cases pairwise (the SPSS default) uses every case that has valid data for the specific pair of variables in that cell, even if the same case is missing data on a different variable elsewhere in the matrix. This maximizes the N used for each individual correlation but means different cells in the same matrix can be based on different, overlapping-but-not-identical subsets of cases.
• Exclude cases listwise drops any case that is missing data on any variable entered into the procedure, so every cell in the matrix is computed on the exact same set of cases and the same N throughout — at the cost of a smaller N overall if missingness is spread across variables.

Pairwise deletion is the more common default for a simple correlation matrix, but listwise deletion is preferable when you plan to compare coefficients across cells directly, since pairwise N differences can make two correlations technically incomparable. Whichever you use, report it — a reviewer cannot tell which was applied just by looking at the table.

Formatting a full matrix for a results table, and the caveat the raw output hides

When more than two variables go into the Variables box, the printed table is a full k × k matrix, but it is also a symmetric one: the cell for (Variable A, Variable B) reports the identical r, p, and N as the cell for (Variable B, Variable A) on the other side of the diagonal, and the diagonal itself is always 1.000 with no significance value. A results table in a paper virtually never reproduces the whole square — the standard convention is to report only the lower (or upper) triangle, omit the diagonal entirely, number the variables (1, 2, 3…) in the row and column headers instead of repeating full variable names, and keep the asterisk convention SPSS already applies (* p < .05, ** p < .01) with a footnote defining it. Rebuilding this by hand from the SPSS output — copying only the lower triangle into a word processor or spreadsheet table — is faster than trying to reformat SPSS’s own pivot table for publication.

The caveat a raw matrix output hides: none of those asterisks are corrected for multiple comparisons. Each cell is an independent significance test run at the nominal α you set (.05 by default), and a matrix built from k variables runs k(k−1)/2 such tests at once — 10 variables means 45 correlations, 15 variables means 105. SPSS does not adjust any of them for having been run together; it flags each pair exactly as it would if that pair were the only test performed. Purely by chance, roughly 5% of pairs with no real underlying relationship will still land under .05, so a 45-cell matrix can show a couple of “significant” asterisks even when nothing in the data is actually related. This matters most for an exploratory matrix — entering every variable in the dataset and scanning for asterisks afterward — and matters far less for a small, theory-driven set of correlations specified before the analysis was run. If the write-up leans on the significance of a specific handful of pairs, consider a Bonferroni adjustment (divide the target α by the number of comparisons actually being interpreted, not necessarily every cell in a large background matrix) or state plainly that the matrix is exploratory and uncorrected — a reviewer who knows the multiple-comparisons problem will ask if the write-up doesn’t address it first.

Writing it up in APA format

The standard reporting line is:

r(df) = [value to two decimal places], p = [exact value, or p < .001 if SPSS rounds it to .000]

For example: r(38) = .45, p = .003. The degrees of freedom for a Pearson correlation are N − 2 (not the N printed in the output table directly) — for a sample of 40 pairs, df = 38, which is what goes in parentheses, not 40. Always name which coefficient was used (“Pearson’s r” or “Pearson product-moment correlation”) rather than the generic word “correlation,” since the assumptions, degrees-of-freedom convention, and interpretation differ from Spearman’s rho or Kendall’s tau. Report the direction (positive or negative) and, where relevant to the argument you are making, treat r or r² as an effect size rather than stopping at significance alone, since fixed verbal labels like “strong” or “moderate” are field-dependent conventions, not universal statistical cutoffs — see the effect size guide for how to choose and interpret one in context.

The SPSS procedure for another common bivariate test, the paired- or independent-samples t-test, follows the same read-the-output-then-write-APA pattern — see the t-test in SPSS guide if that’s the next analysis on your list. For a categorical-variable equivalent, the chi-square test entry has its own SPSS walkthrough (Crosstabs, not Bivariate Correlations).

Frequently asked questions

What does “Sig. (2-tailed)” mean in the SPSS correlation output?

It is the p-value for the null hypothesis that the true (population) correlation is zero, tested against both directions of association. It tells you whether the observed r is distinguishable from zero given your sample size — it does not by itself tell you whether the relationship is strong or practically important.

Why is N different in different cells of my correlation matrix?

Because Exclude cases pairwise (SPSS’s default missing-value handling) computes each cell from whichever cases have valid data on that specific pair of variables, so a variable with more missing values will show a lower N wherever it appears in the matrix. Switch to Exclude cases listwise under Options if you need one consistent N across every cell.

How do I check for outliers before running a Pearson correlation in SPSS?

Produce a scatterplot of the two variables first (Graphs > Chart Builder, Scatter/Dot), before running the Bivariate Correlations procedure. Look for points well away from the main cloud of data, particularly toward the extremes of either axis, since these have the most leverage on the coefficient.

Should I report r or r²?

Report r as the primary statistic, since it preserves the direction of the relationship; r² (always positive) is commonly reported alongside it as the proportion of shared variance when that framing is relevant to your argument, but it discards the direction information r carries on its own.

Does SPSS correct a correlation matrix for multiple comparisons?

No. Every cell in a Bivariate Correlations matrix is tested independently at the significance threshold you set, with no adjustment for how many pairs are being tested at once. A matrix built from k variables runs k(k−1)/2 separate tests, so a handful of “significant” asterisks can appear by chance alone in a large, exploratory matrix. Apply a manual correction, such as Bonferroni (divide the target α by the number of comparisons being interpreted), or limit the matrix to a small, theory-driven set of pairs, if the write-up depends on which specific correlations reached significance.

What’s the difference between the Bivariate and Partial Correlations procedures in SPSS?

Analyze > Correlate > Bivariate gives the raw, two-variable correlation with nothing controlled for. Analyze > Correlate > Partial gives the correlation between two variables after statistically removing the linear effect of one or more additional control variables — use it when a third variable is a plausible confound for the association you’re testing.

For the descriptive statistics (means, SDs, distribution shape) you’d typically run before a correlation to understand each variable on its own, see the descriptive statistics guide. For more SPSS procedure walkthroughs, browse the research tools hub.